| 29 Jun 2025 |
 emily | maintaining that long-term is another thing | 02:50:23 |
| emily | but it is easier to sell the initial switchover without it at least (see: that PR…) | 02:50:33 |
| emily | * but it is easier to sell the initial switchover with it at least (see: that PR…) | 02:50:37 |
 hexa | https://github.com/NixOS/nixpkgs/pull/287043#issuecomment-1933060123 | 02:50:39 |
| hexa | olne month to migrate network configurations, tough | 02:50:55 |
| hexa | can't be done | 02:50:56 |
| hexa | * one month to migrate network configurations, tough | 02:51:30 |
| emily | anyway systemd stage 1 is definitely something that needs to happen | 02:51:30 |
| emily | I know ElvishJerricco has a list of the few remaining blockers | 02:51:42 |
| emily | if anyone has time to help solve them it would make it more likely to happen for 25.11 I'm sure | 02:52:01 |
| hexa | https://github.com/NixOS/nixpkgs/issues/251388 😱 | 02:52:50 |
| hexa | I hate everything about that issue | 02:53:12 |
| emily | "The networking.* options aren't fully supported in scripted networking + initrd anyway." :) | 02:54:03 |
| emily | I think that can be WONTFIX | 02:54:12 |
| emily | there's definitely a lot of NixOS that originates from a time before there were any real abstractions in the underlying system tooling | 02:54:43 |
| emily | leading to a bunch of Nix with bespoke interfaces hard-wired up to a pile of Bash spaghetti | 02:55:08 |
| emily | RFC 42 comes for us all | 02:55:24 |
| hexa | exactly | 02:55:38 |
| emily | I remember how magical the idea of NixOS seemed back before systemd added declarative configuration for everything in the world though :) | 02:56:51 |
 antifuchs | (I’d love to use systemd-initrd with this machine but it lacks all sorts of features that would make it viable with my current configuration, most of all a tpm2 so transferring secrets over even if they’re not all that secret is annoying; also, anytime I mess up I have to do tedious work to recover it via a virtual kvm that sucks, across an ocean… so whenever systemd-initrd turns into the only option that machine gets retired; until then, I script harder) | 03:19:41 |
 ElvishJerricco | I don't believe useNetworkd by default depends on systemd initrd in any way | 03:24:35 |
| ElvishJerricco | the old networking in scripted initrd will be fine with it | 03:24:48 |
| ElvishJerricco | What can systemd initrd not do that you need? | 03:25:29 |
| ElvishJerricco | AFAIK the only major feature it lacks is the old LUKS nonsense, but the new LUKS stuff is better | 03:25:41 |
| ElvishJerricco | like the tpm2 is not needed to do any secrets stuff; boot.initrd.secrets is effectively implemented the same way in both scripted in systemd initrd | 03:27:06 |
| antifuchs | In reply to @elvishjerricco:matrix.org
What can systemd initrd not do that you need? It’s mainly a problem with some modules I wrote: they expect encrypted systemd credentials for sysd-initrd; falling back on initrd secrets in scripted | 03:27:21 |
| ElvishJerricco | then I don't understand what the problem is | 03:27:58 |
| antifuchs | Yeah it’s not necessary but they still end up in the nix store decrypted, which id like to avoid | 03:28:00 |
| ElvishJerricco | boot.initrd.secrets does not necessarily put secrets in the store | 03:28:17 |
| antifuchs | Indeed, it’s my module. And that needs a rewrite and I’m not eager to do that because any change there can break the machine and that means a wasted afternoon of vnc recovery | 03:29:19 |
